Abstract
The vibration signals of rolling bearings are often nonlinear and non-stationary. Multiscale entropy (MSE) has been widely applied to measure the complexity of nonlinear mechanical vibration signals, however, at present only the single channel vibration signals are used for fault diagnosis by many scholars. In this paper multiscale entropy in multivariate framework, i.e., multivariate multiscale entropy (MMSE) is introduced to machinery fault diagnosis to improve the efficiency of fault identification as much as possible through using multi-channel vibration information. MMSE evaluates the multivariate complexity of synchronous multi-channel data and is an effective method for measuring complexity and mutual nonlinear dynamic relationship, but its statistical stability is poor. Refined composite multivariate multiscale fuzzy entropy (RCMMFE) was developed to overcome the problems existing in MMSE and was compared with MSE, multiscale fuzzy entropy, MMSE and multivariate multiscale fuzzy entropy by analyzing simulation data. Finally, a new fault diagnosis method for rolling bearing was proposed based on RCMMFE for fault feature extraction, Laplacian score and particle swarm optimization support vector machine (PSO-SVM) for automatic fault mode identification. The proposed method was compared with the existing methods by analyzing experimental data analysis and the results indicate its effectiveness and superiority.
Highlights
At present most mechanical fault diagnosis methods are constructed based on the vibration signals collected from single channel or direction, while the vibration information signals of other channels or directions are often ignored
Many nonlinear dynamics methods including approximate entropy (AppEn) [6,7], sample entropy (SampEn) [8,9] and Multiscale entropy (MSE) [10,11,12] have been widely applied to mechanical fault diagnosis due to their ability to effectively extract the fault feature information hidden in the vibration signals that cannot be well extracted by linear analysis methods
The comparison result indicate that the multiscale fuzzy entropy (MFE), multivariate multiscale fuzzy entropy (MMFE)- and Refined composite multivariate multiscale fuzzy entropy (RCMMFE)-based methods can reflect the fault information better and get higher identifying rates than the MSE- and multivariate multiscale entropy (MMSE)-based ones no matter whether using Laplacian score (LS) is used for feature selection
Summary
At present most mechanical fault diagnosis methods are constructed based on the vibration signals collected from single channel or direction, while the vibration information signals of other channels or directions are often ignored. Many nonlinear dynamics methods including approximate entropy (AppEn) [6,7], sample entropy (SampEn) [8,9] and MSE [10,11,12] have been widely applied to mechanical fault diagnosis due to their ability to effectively extract the fault feature information hidden in the vibration signals that cannot be well extracted by linear analysis methods. The refined composite multivariate generalized multiscale fuzzy entropy (RCMvGMFE) was proposed by Azami and Escudero [20] to improve the performance of MMFE in complexity analysis of multi-channel signals.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
